Toshiba America Electronic Components Inc v. Monument Peak Ventures LLC

1. Case Identification

• Case #: IPR2021-00330
• Patent #: 7,583,294
• Filed: December 16, 2020
• Petitioner(s): Toshiba America Electronic Components, Inc.
• Patent Owner(s): Monument Peak Ventures, LLC
• Challenged Claims: 1, 5, 10, 11, and 19

2. Patent Overview

• Title: Method for Detecting Faces in Images
• Brief Description: The ’294 patent relates to image processing methods for detecting human faces within a digital image. The core technology is a two-stage algorithmic scheme where a first, faster algorithm prescreens image data to identify potential face candidates, and a second, more sophisticated algorithm processes those candidates to confirm the presence of a face with a lower rate of false positives than the first algorithm.

3. Grounds for Unpatentability

Ground 1: Anticipation by Yow - Claims 1, 5, 10, and 19 are anticipated by Yow.

• Prior Art Relied Upon: Yow (K.C. Yow & R. Cipolla, Feature-Based Human Face Detection, Image and Vision Computing, 1997).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner argued Yow discloses a two-stage, feature-based algorithm for face detection. The first stage, a "perceptual grouping framework," prescreens image data to find "interest points" and group them into face candidates using geometric and gray-level information, which Petitioner mapped to the "pattern matching technique based on color and shape information" of claim 1. The second stage uses a "probabilistic framework" with a Bayesian classifier to analyze the candidates and confirm faces, which Petitioner asserted meets the "posterior probability function classifier" limitation. Petitioner contended that Yow's first stage produces a "reasonable number of false positive faces" which the second stage is designed to eliminate, thus disclosing a second algorithm with a lower rate of false positives.
  • Key Aspects: Petitioner argued that Yow’s disclosure of superimposing rectangular frames and symbols over detected faces meets the dependent claim limitation of providing an indication of detection.

Ground 2: Anticipation by Kresch - Claims 5 and 10 are anticipated by Kresch.

• Prior Art Relied Upon: Kresch (Patent 6,463,163).
• Core Argument for this Ground:
  • Prior Art Mapping: Petitioner contended that Kresch discloses a two-step method for face detection. The first step uses a "candidate selector" that pre-filters an image to select regions potentially containing a face, thus identifying a plurality of face candidates. The second step employs a "face detector" (e.g., a neural network) to analyze these candidates and verify if they contain a face. Petitioner asserted that Kresch explicitly teaches that the face detector may reject candidates selected by the first step, thereby demonstrating that the second algorithm has a lower rate of false positives than the first. The dependent claim limitation of providing an indication was allegedly met by Kresch's disclosure of producing an output signal or superimposing an outline on the image.

Ground 3: Obviousness over Yow - Claims 1, 5, 10, 11, and 19 are obvious over Yow in view of a POSA’s Knowledge.

• Prior Art Relied Upon: Yow (K.C. Yow & R. Cipolla, Feature-Based Human Face Detection, Image and Vision Computing, 1997) and the general knowledge of a Person of Ordinary Skill in the Art (POSA).

• Core Argument for this Ground:

  • Prior Art Mapping: This ground serves as an alternative to the anticipation argument under Ground 1. Petitioner argued that to the extent any limitation is not explicitly disclosed in Yow, it would have been obvious to a POSA to modify Yow’s system to arrive at the claimed invention. For example, if Yow’s use of geometric and gray-level information was not considered to meet the "color and shape" limitation, Petitioner argued it would have been obvious to use well-known color information in the initial screening stage.
  • Motivation to Combine: A POSA would combine Yow's teachings with general knowledge of image processing to improve the efficiency and robustness of the face detection system. Implementing a second, more rigorous stage to reduce false positives from a faster initial stage was presented as a matter of common sense and a common design choice in the field to balance speed and accuracy.
  • Expectation of Success: A POSA would have had a high expectation of success because implementing a two-stage filtering process where the second stage is more selective than the first is a predictable method for improving detection accuracy.

• Additional Grounds: Petitioner asserted additional anticipation challenges against claims 5 and 10 based on Chang (WO 99/23600) and Yang (a 1994 journal article). Petitioner also asserted additional obviousness challenges based on combinations of Kresch, Chang, and Yang, each in view of a POSA's knowledge, relying on similar common-sense design modification theories.

4. Key Claim Construction Positions

• "false positives": Petitioner proposed this term be construed as "incorrect indications that a face is present." This construction was argued to be consistent with the patent's goal of having the second algorithm "eliminate the many false [face] detections that pass through" the first.
• "algorithm interface adapter": Petitioner proposed the construction "a component that allows implementation of an algorithm." This was based on the claim language reciting prescreening "using" the adapter, and the specification's depiction of it as a functional block connected to a computer bus without further structural definition.
• "posterior probability function classifier": Petitioner proposed construing this term as "a Bayesian classifier that performs maximum posterior classification using a stored probability distribution." This was argued to be mandated by the specification’s repeated references to Bayes' theorem as the underlying principle for this classifier.

5. Arguments Regarding Discretionary Denial

• Petitioner argued that discretionary denial under Fintiv would be improper. The parallel district court litigation was terminated via a final judgment of patent ineligibility under 35 U.S.C. §101 before significant investment occurred. Consequently, Petitioner asserted that there is no trial date, the parties' investment in the parallel litigation was minimal, and there is no overlap between the §101 issue on appeal and the anticipation and obviousness issues presented in the IPR. Petitioner also argued that since none of the asserted prior art was before the examiner during prosecution, institution would advance the mission of improving patent quality.

6. Relief Requested

• Petitioner requests institution of IPR and cancellation of claims 1, 5, 10, 11, and 19 as unpatentable.